[ClusterLabs] Is 20 seconds to complete redis switchover to be expected?

[steven prothero]

Hello,

UPDATE on this situation is good news!

I edited some items in corosync.conf  while reading:
https://manpages.debian.org/testing/corosync/corosync.conf.5.en.html

token: 3000 → 1000
token_retransmits_before_loss_const: 10 → 4
join: 60 → 50
consensus: 3600 → 1200
expected_votes: 2 →   # expected_votes: 2  (commented out)

So timing is improved:

:01 - node1 crash (redis-server killed)
:01 - Pacemaker knows there is a problem.
:02 - STONITH starts.
:08 - STONITH finished.
:12 - Redis switched.
:13 - VIP switched.

After stonith finished, chit-chat went from 10 seconds →  4 seconds....

1. I will investigate version of fence_ipmilan for ways to improve.
2. I will look at the new logs for what is happening after the
stonith/fence is finished and Redis is switched. (the chit chat)

Steven